1. Chapter 2.
Characterising the Next Generation of Mobile
Applications Through a Privacy-Aware
Geographic Knowledge Discovery Process
- M. Wachowicz, A. Ligtenberg, C. Renso, and S. G¨urses
2. Index
2.5 The Multi-Tier Ontological Framework
2.5.1 Tier 0: The Reality Space
2.5.2 Tier 1: The Positioning Space
2.5.3 Tier 2: The Geographic Space
2.5.3 Tier 3: The Social Space
2.5.4 Tier 4: The Cognitive Space
2.6 Future Application Domains for a Privacy-Aware GKDD Process
2.6.1 Transport Management: The Integration of Multimodal Choices
2.6.1.1 Tier 0: The Reality Space
2.6.1.2 Tier 1: The Positioning Space
2.6.1.3 Tier 2: The Geographic Space
2.6.1.4 Tier 3: The Social Space
2.6.2 Spatial Planning: The Adaptation of Space to Human Behaviour
2.6.2.1 Tier 0: The Reality Space
2.6.2.2 Tier 1: The Positioning Space
2.6.2.3 Tier 2: The Geographic Space
2.6.2.4 Tier 3: The Social Space
2.6.3 Marketing: The Shift Towards Movement-Aware Marketing
2.6.3.1 Tier 0: The Reality Space
2.6.3.2 Tier 1: The Positioning Space
2.6.3.3 Tier 2: The Geographic Space
2.6.3.4 Tier 3: The Social Space
2
3. 2.5 The Multi-Tier Ontological Framework
• A GKDD process constructed from a multi-tier ontological perspectiv
e aims to integrate different reasoning tasks in a unified system by
mapping the complex relationship between movement metaphors an
d patterns.
– This will lead to uncovering new and innovative hypothesis of distributions, patter
ns and structures across very large databases.
– Therefore, these metaphors will not rely on similar reasoning backgrounds but wil
l be derived from the integration of different inference modes (i.e. abductive, indu
ctive, and deductive).
3
4. 2.5 The Multi-Tier Ontological Framework
Known Tier 0 : Reality Space
Metaphors A-priori Knowledge
Sample
Tier 1 : Positioning Space
Observations Deductive Reasoning (What, Where, When)
Relation
Tier 2 : Geographic Space
Confront Context Deductive Reasoning (How, What for)
Discovery
Tier 3 : Social Space
Model Inductive Reasoning (What if)
Insight
Tier 4 : Cognitive Space
Discovered Cognitive Tacit Knowledge
Metaphor Abductive Reasoning (Why)
4
5. 2.5.1 Tier 0: The Reality Space
• Tier 0 of the ontology represents the ‘reality space’, which recognise
s the existence of a known world as a four-dimensional continuous fi
eld in space and time.
– Several known movement metaphors are currently being used, including the mov
ement-as-journey metaphor and the movement-as-activity
movement-as-journey
• one or two journeys on a day are most common
• over three quarters of all(75%) journeys are usually a single-stop tour
• while combining more than three stops in a journey is very rare.
movement-as-activity
• More than 50% of all out-of-home activities take more than an hour and over 30% take eve
n more than 2h [3].
5
6. 2.5.1 Tier 0: The Reality Space
• In tier 0, it is important to establish whether there are any privacy co
ncerns from any of the sensor carriers and the experts of an applicat
ion domain.
– define a level of privacy according to who are the sensor carriers from whom dat
a will be collected and who are the involved experts who will define the purposes
of collecting these data.
– Once the stakeholders are identified in tier 0, it is also necessary to identify what
their privacy requirements are, which could there be stated in terms of hierarchic
al levels of privacy.
6
7. 2.5.2 Tier 1: The Positioning Space
Known Tier 0 : Reality Space
Metaphors A-priori Knowledge
Sample
Tier 1 : Positioning Space
Observations Deductive Reasoning (What, Where, When)
Observations
• measurement values at every point in space and time,
• based on some measurement scale, which may be quantitative or qualitative.
• always marked by some degree of uncertainty, which depends on the type of sensors bein
g used for collecting the location and movement information of mobile entities, such as X,Y
,Z coordinates, speed and time.
• They can be navigation sensors (e.g. GPS, INS, MEMS sensors, digital compasses, etc.),
remote sensing sensors (e.g. frame-based cameras, thermal cameras, laser scanners, etc
.) and wireless technologies.
7
8. 2.5.2 Tier 1: The Positioning Space
• We do not have a trajectory representation of these points yet.
• However, it is possible to distinguish the observations according to f
our point representations.
• Stop : A cluster of points that represent stops with a very short duration of some minutes d
ue to traffic light or stop signs.
• Stop over : A cluster of points that represent a change of speed. For example, a road accid
ent.
• Short stay : A cluster of points that represent stays with a short duration of some hours du
e to an activity such as working, shopping or leisure.
• Long stay : As cluster of points that represent stays with a long duration of several hours t
hat will correspond to the sensor device being switch off or being at home.
8
9. 2.5.2 Tier 1: The Positioning Space
• The reasoning task consists of allowing one to infer a consequence f
rom a set of point patterns.
– The consequences are drawn from these point patterns down to a specific metap
hor such as, for example, the movement-as-urban forms metaphor.
(1) Ring network in a concentric city: concentric pattern
(2) Radial network in a lob city: radial pattern
(3) Linear poly-nuclear city: linear concentric pattern
(4) Concentric poly-nuclear city: circular concentric pattern
(5) Linear network in a linear city: linear pattern
(6) Grid city: square concentric pattern
9
11. 2.5.2 Tier 1: The Positioning Space
• In terms of privacy,
– the main issue is to make sure that the granularity of data collection is in accorda
nce with the privacy requirements of the sensor carriers.
– The domain experts can also make sure that the granularity of positioning data s
et is appropriate for the needs of the application domain and it complies with the
privacy requirements of the different sensor carriers.
11
12. 2.5.3 Tier 2: The Geographic Space
Tier 1 : Positioning Space
Observations Deductive Reasoning (What, Where, When)
Relation
Tier 2 : Geographic Space
Context Deductive Reasoning (How, What for)
• From the movement metaphors and representations already defined
in the previous tier 1, a trajectory is defined in this tier as any polylin
e between stops, stop overs, short stays and/or long stays.
• Moreover, in this tier is where the privacy requirements of the sensor
carriers need to be implemented using security constraints.
– If there are sensor carriers who want their trajectories to be unobservable or who
want to remain anonymous, then the necessary steps need to be taken here by a
pplying different trajectory privacy-preserving methods such as cloaking and mix
es.
12
13. 2.5.3 Tier 2: The Geographic Space
• Cloaking
• Mix
13
14. 2.5.3 Tier 2: The Geographic Space
• Currently, the main metaphor being used at this ontological level is a
ccessibility, which can be obtained by the calculation of trajectory di
stances (lengths).
• However, the GKDD process is entirely propositional and different m
ovement metaphors need to be taken under consideration at this ont
ological level.
– the metaphors of movement-as-urban form and movement-as-accessibility can b
e used to deduce the consequences of the linear patterns,
• from the trajectories based on the generalization of the trajectories
• using the transportation networks such as the types of streets in an urban area.
14
15. 2.5.3 Tier 2: The Geographic Space
• Accessibility is constrained by urban forms such as the transportatio
n network.
– the tree street patterns usually impede the movement of people on reaching a de
stination.
– grid street patterns facilitate the fast reaching to a particular destination.
• A GKDD process might provide the data mining query mechanism n
ecessary to discover an anomaly in the trajectories
15
16. 2.5.4 Tier 3: The Social Space
Tier 2 : Geographic Space
Context Deductive Reasoning (How, What for)
Discovery
Tier 3 : Social Space
Model Inductive Reasoning (What if)
• The tier 3 encompasses the model that underlies our daily trajectorie
s and their fundamental relations with human activities.
– transportation modalities (e.g. car or public transport),
– commuting time or distance,
– spatial distribution of jobs and housing locations,
– total vehicle miles travelled,
– average trip lengths and congestion on links and intersections.
– from socio-economic statistics, such as income and education of a neig
hbourhood.
16
17. 2.5.4 Tier 3: The Social Space
• It is important to realize that a trajectory takes place in a social-spati
al system.
social system
• consists of individuals, groups and organizations that maintain relations through intentional (
cooperative) activities,
• based upon a more or less common set of rules, norms and values, and acts within the bou
ndaries of the institutions that are derived from it [13]
spatial system
• is composed of biotic and abiotic components, processes that alter these components and r
elations between them.
• The main metaphor used at this tier level is movement-as-activity.
– The spatial system and the social system are strongly intertwined and should not
be analyzed separately.
17
18. 2.5.4 Tier 3: The Social Space
• In this tier, it requires that some sort of target must already have bee
n identified, and the task becomes one of uncovering ‘what if’ scenar
ios to explain the trajectory patterns within a social context.
– For example, instead of finding the best location for a supermarket based on the
proximity of objects on the landscape, the problem becomes about finding the be
st location based on the patterns of the trajectories of people, which in turn, sugg
est that the human activity on a landscape is potentially more complex and proba
bilistic.
• Some examples are given below:
– Discover patterns that explain the occurrence of a certain activity (e.g. shopping,
recreation)
– Discover the dependencies between different characteristics of activities
– Discover the activities subsets and time periods with the corresponding patterns
– Detect the occurrence of an unexpected activity
18
19. 2.5.4 Tier 3: The Social Space
• The data miner and the domain expert need to make sure that exact
rules that contain the complete data population do not breach any of
the ‘group privacy’ requirements.
• there may be privacy requirements in the following form:
– If the size of an identified group is less than 10% of the complete population, the
n a sensor carrier wants to be unobservable, or it should not be inferred that the
sensor carrier belong to this group.
– If an unexpected activity(example number 4 above) is detected, and this contains
information about clearly identifiable locations, small set of trajectories or small gr
oups of people, then this information should either not be released or only acces
sible to trusted parties.
19
20. 2.5.5 Tier 4: The Cognitive Space
Tier 3 : Social Space
Model Inductive Reasoning (What if)
Insight
Tier 4 : Cognitive Space
Discovered Cognitive Tacit Knowledge
Metaphor Abductive Reasoning (Why)
• the goal of a geographic knowledge discovery process is to gain kno
wledge through abductive reasoning that can function in the absenc
e of pre-determined hypotheses, training examples or rules
• it is important to point out the difference between tacit and implicit kn
owledge.
– implicit knowledge is something experts might know, but not wish to express
– tacit knowledge is something that experts know but cannot express, it is personal
, difficult to convey, and which does not easily express itself in the formality of lan
guage.
20
21. 2.5 The Multi-Tier Ontological Framework
Known Tier 0 : Reality Space
Metaphors A-priori Knowledge
Sample
Tier 1 : Positioning Space
Observations Deductive Reasoning (What, Where, When)
Relation
Tier 2 : Geographic Space
Confront Context Deductive Reasoning (How, What for)
Discovery
Tier 3 : Social Space
Model Inductive Reasoning (What if)
Insight
Tier 4 : Cognitive Space
Discovered Cognitive Tacit Knowledge
Metaphor Abductive Reasoning (Why)
21
22. 2.6 Future Application Domains for
a Privacy-Aware GKDD Process
• Three application domains have been selected to illustrate the expe
cted innovations on applications in,
– transport management
– spatial planning
– marketing
http://www.microlise-america.com/index.php/portfolio/transport-management-centre
http://www.flickr.com/photos/38702466@N05/3563816815/
http://webbusiness2go.com.au/location-based-marketing/ 22
23. 2.6.1 Transport Management:
The Integration of Multimodal Choices
• Transport or transportation refers to the movement of people and go
ods from one place to another.
• Transport management is aimed at solving the problems between inf
rastructure (e.g. transport networks) and operations (e.g. road traffic
control).
23
24. 2.6.1 Transport Management:
The Integration of Multimodal Choices
Known Tier 0 : Reality Space
• travellers and planners with multimodal information about their trajectory behaviour.
Metaphors
Sample
Observations Tier 1 : Positioning Space
• Most common metaphors : movement-as-accessibility
• that defines how people move from one location to another as pedestrians, or takin
g cars, bikes or public transportation.
• For a geographic knowledge discovery process, this might imply the deductive sea
rch as one of the following examples:
1. Discover how a set of point patterns evolves from time t1 to time t2, in terms of a s
pecific mode of transportation
2. Discover an observation window (spatial and temporal extends) where point patte
rns reveal a change of mode of transportation
3. Discover the rules that explain a spatial distribution of a set of point patterns at a
given time in terms of a specific mode of transportation
• The domain experts need to consider if any of these discovered patterns are in co
nflict with the privacy or security requirements of any of the sensor carriers.
24
25. 2.6.1 Transport Management:
The Integration of Multimodal Choices
Relation
Tier 2 : Geographic Space
Context • patterns of moving people and their respective trajectories will have an impact on un
derstanding the relationship between modality choices and trajectory patterns.
• the knowledge of such trajectory patterns will play an important role in studying the r
oute conditions from effects such as hazards, noise, traffic jams and visual pollution.
• This kind of information would allow the experts to check if information can be inferre
d from the trajectory representation that may breach the privacy requirements of any
Discovery of the sensor carriers.
Tier 3 : Social Space
Model • Some examples are given below:
1. Discover accessibility patterns that explain the occurrence of shopping activity with
its corresponding transportation modality
2. Discover the dependencies between working and leisure activities according to a s
pecific transportation modality
Insight 3. Detect the occurrence of an unexpected activity
• the data miners and domain experts need to observe the classifications that they inf
er and check to see whether the ‘category anonymisation’ requirements of the differ
ent sensor carriers are breached.
25
26. 2.6.2 Spatial Planning:
The Adaptation of Space to Human Behavior
• Spatial planning is aimed to change the organisation of a geographi
c environment to meet the demands of society.
– Demands of society continuously change as the result of change in the society a
nd also due to change in the geographic environment itself.
– As space becomes a limited resource the geographic environment is expected to
fulfil multiple functions [54].
• In spatial planning,
– location-allocation representations and methods have been developed when posi
tioning data sets were scarce and difficult to obtain, and models were determinist
ic or entirely predictable.
– In principle, mobile technologies made it possible to gather very large data sets c
ontaining movement information from mobile devices over time.
– This has opened the opportunity to deal with the location-allocation problems fro
m a people’s perspective in spatial planning.
26
27. 2.6.2 Spatial Planning:
The Adaptation of Space to Human Behavior
Known Tier 0 : Reality Space
Metaphors • land use could become the activity metaphor based on the movement of people, rath
er than the location of its features.
• Knowledge of the movement of people in these types of areas is required for the situ
ating of shops, shop-types and checkpoints.
• More over the dimensioning of pathways, gateways and emergency evacuation route
Sample s might benefit from additional knowledge about the spatial temporal dynamics of mo
ving crowds.
Tier 1 : Positioning Space
Observations • The above mentioned examples of applications in spatial planning would benefit from
the gathering of positioning data of moving people on a landscape into a trajectory da
ta warehouse.
• This type of data neither is commonly used in the process of designing spaces nor is
it commonly available.
27
28. 2.6.2 Spatial Planning:
The Adaptation of Space to Human Behavior
Relation
Tier 2 : Geographic Space
Context • The geographic discovery process needs to be essentially targeted to finding linear p
atterns of trajectories that can be understood by a planner and designer.
• The main metaphor at this tier is movement-as-urban form.
• certain land use type might enable common activities to occur close to a specific pla
ce (e.g. housing and food shopping), as well as places with higher density developm
ent closer to transportation lines and hubs. Poor land use concentrates activities (su
Discovery ch as jobs) far from other destinations (such as housing and shopping).
Tier 3 : Social Space
Model • Multifunctional land use can be defined as combining various socio-economic activiti
es in the same area.
• The challenging factor is to provide ample opportunities for recreational activities whi
le preserving and developing nature.
• Knowledge about the patterns of movement of visitors of nature areas and the type o
Insight f activities might improve the harmonisation of multifunctional use of nature areas.
• Individuals may not want their leisure activities to be so clearly identifiable.
28
29. 2.6.3 Marketing:
The Shift Towards Movement-Aware Marketing
• Currently marketing is mostly done based on customers’ profile, whi
ch, normally, is statically defined.
• Traditionally, the success of marketing depends on what is called th
e marketing mix of four P’s: product, price, promotion and placement
.
– This rather traditional view on marketing has been criticised, since its main focus
is on a company or marketer rather than the consumer.
• Geo-marketing implies the use of GIS to add location information int
o the marketing mix.
– Based on spatial analysis, additional knowledge and insight might be gained abo
ut, for example, the spatial distribution of income and demographic composition o
f districts.
• The main metaphor is movement-as-personalisation.
– many people to be spammed by location-based advertisements, generated by rel
ative dumb LBS, can be alleviated by providing more intelligent information base
d on movement behaviour.
29
30. 2.6.3 Marketing:
The Shift Towards Movement-Aware Marketing
Known Tier 0 : Reality Space
Metaphors • Using LBS, marketers can pinpoint their marketing mix and enhance their communic
ation with potential customers based on their exact location and time.
• movement-based services (MBS)
• LBS only provide the context from the users, and the environment (who is where at ti
me t). MBS have the potential to add to this, knowledge about what he/she did, how
Sample he/she did it and with whom.
Tier 1 : Positioning Space
Observations • LBS usually describe only the location in space, a caller id and a time stamp.
• MBS, information about the followed tracks, the movement characteristics (speed, a
cceleration, periodicity in movements, etc.) need to be added and stored.
• MBS typically require the maintenance and storage of data to be able to infer patter
ns out of it.
• As data are required and requesting of the movements of individual people preservi
ng privacy is an important requirement.
• in marketing-based application, the control of the right should be part of the decision
making about what part of the reality space should be sampled and registered.
30
31. 2.6.3 Marketing:
The Shift Towards Movement-Aware Marketing
Relation
Tier 2 : Geographic Space
Context • marketing-related behaviour and movement behaviour.
1. the consent model is based on informing or assisting users with information or servi
ces based on authorisation given by them.
2. users receive targeted information based on their movement behaviour, location, ti
me and the behaviour of others
• privacy in such an application depends on which of the above location information t
Discovery he sensor carrier is unwilling to have analysed.
Tier 3 : Social Space
Model • The discovery of geographic knowledge related to privacy aware marketing is mainly
targeted to finding groups that show similar behaviour and to determine if this behavi
our is interesting, given a certain marketing goal.
1. Discover the general patterns that explain the behaviour of certain groups of people
given a marketing perspective.
Insight 2. Discover the dependencies between movement behaviour and the effects of perso
nalised movement-aware marketing.
3. Discover the type of information appreciated by people when they are moving at a c
ertain time, modality and location.
31
Hinweis der Redaktion
지식발견 과정은 쉬운 과정이 아니며, 거기서 사용되는 메타포들의 특성, 정보들간의 유사성과 차이성, 상호관계, 행동등 다양한 면을 검토해야 한다.Our aim is to describe a GKDD process using ontological tiers that will provide the common base for the organisation of different nature and sources of knowledge of the movement metaphors used by experts of application domains. The tiers also establish the movement metaphors for the integration of different reasoning tasks in a unified system. This section describes the multi-tier ontological framework that has been developed from two previous fundamental research works: First, the work on a set of tiers of ontology previously proposed by Frank [15] for defining consistency constraints, data interoperability and more recently data quality in Geographical Information Systems (GIS) [33] Second, our multi-tier framework largely based on the three ‘spaces’ paradigm that has been proposed by Ernst Cassirer(1874–1945)
which are named sampling, relating to a geographic context, discovering patterns, generating new insights and confronting them with previous background knowledge. From a privacy perspective, the multi-tier framework allows a number of legal frameworks, often specific to application domains, to be adhered to throughout the GKDD process. Often such frameworks require the sensor carriers and domain experts to get the consent of those whose data is being collected to the primary or secondary use of that data. This is particularly challenging in a GKDD process, since the metaphors and the context of their use may not yet be determined during the tier 1 when the collection of data is carried out. Furthermore, the other tiers may also have privacy constraints on the results of the GKDD process that are unknown by the data miner as well as the expert of an application domain.
The process of geographic knowledge discovery may use this type of knowledge for generating a priori knowledge as pre-determined hypothesis, training examples or rules.
In the case of applications for transportation management, the sensor carriers might be those traveling from home to work, and the experts might be the company managers who have privacy goals towards the collected data. Company managers may not want it to be known where their employees travel during work hours, since this could point out to information about the activities of that company and those who are interested in using the data, for example, the supermarkets in the area may be interested in the trajectories relevant for better advertising. 예를 들어 수송관리 업무 관점에서 보면, 회사 관리자(회사 직원의 이동 정보를 수집하는) 입장에서는 회사 직원들의 이동 정보가 외부에 알려지는 것이 회사의 활동이나 관심분야에 관한 정보를 노출시킬 위험이 있다고 판단할 수 있는 반면, 그 범주에 포함된 판매업자들은 회사 직원의 이동 정보를 이용하여 보다 효과적인 광고를 할 수 있을 것이다.
In this tier, the movement metaphors can be used to infer some empirical knowledge from the discovered patterns, such as density clusters of points in space.
In the Real-Time Graz experiment in Austria, observations of cellphone usage have been collected through the city based on a location system where the movement of the cellphones was recorded and tracked with the agreement of the customers [42].In this example, it is possible to visually identify (4) the circular concentric patterns representing possibly a concentric poly-nuclear city, with some (3)linear poly-nuclear patterns as well. It is also important to point out that in this example, the trajectory representation does not exist yet.
granularity of data collection 은 데이터 수집에 대한 상대적 조건? 즉, 센서 케리어의 프라이버시 요구사항에 따라 데이터 수집이 이뤄져야 한다는 말.
The deductive reasoning task is characterized by inferring descriptive knowledge such as, the trajectory characteristics (e.g. space and time), the geographic environment where the trajectory occurs (i.e. landscape), the topological relations between the trajectories and the association between the trajectories and specific features of a landscape. The overall goal is to help the experts to deduce the consequences for the existence of linear patterns of the movement of the trajectories. A set of movement metaphors is necessary to be defined by using some kind of classification scheme, set of association rules or clustering.
Depending on the size of zones, the actual trajectory distance may be significantly different to the distance calculated using average centroid distances [4, 50, 53]. The calculations also do not account for the configuration of the transport network in order to establish the actual route distances. In fact, they are only based on straight line distances between origin and destination zones [51]. Accessibility is constrained by urban forms such as the transportation network. the tree street patterns usually impede the movement of people on reaching a destination. grid street patterns facilitate the fast reaching to a particular destination.
A GKDD process might provide the data mining query mechanism necessary to discover an anomaly in the trajectories corresponding to different types of local streets, the similarities and dissimilarities among the trajectories according to the different characteristics of street types and finally discover point clusters of non-movement among trajectories and their association with the type of local street.
Traditional spatial planning theory usually considers the geographic environment.the national government : the cities, main infrastructure, population densities, nature areas, etc., flows of people and goods. Municipalities : need detailed, high level information about the individual functions of the buildings, detail infrastructure and social compositions of different neighbourhoods.
the planners use various metaphors for the geographic environment, depending on the context of the spatial planning. currently based on mostly static models of activities.the same geographic environment should also be considered as the result of movement patterns of people represented by their invisible footprints of trajectories on the landscape. Pulselli [39] has already pointed out that although positioning data sets of mobile entities are becoming increasingly available, surprisingly enough, they have not been used to describe the social and spatial systems.
The tier 3 is where the ‘classification anonymity’ or the ‘categorization anonymity’ requirements of the sensor carriers need to be guaranteed.
Searle [46] argues that cognitive tacit knowledge is not a form of knowledge (such as beliefs, theories and empirical hypothesis) but rather the preconditions of forming an individual’s background knowledge. This raises the possibility that at least some metaphors of background knowledge can be confronted with the ones of cognitive tacit knowledge, which implicates that the features of the world are not independent of the mind.
for example a business in a remote area, the traffic in and out of the building may be identifiable. If this information is made public, it may allow competing businesses to analyse the traffic of the remotely located business, allowing them to make inferences about their activities. A related example actually is in the US, where Bill of Lading and Ship Manifests collected in a system called ‘US Customs Automated Manifest Systems’ that include information about the trajectories of ships, the origin and the target organisations are made public in adherence to the ‘Freedom of Information Act.’ As a result, competing companies in the US are able to profile the import/export activities of many European companies with the Americas [45].